Many of the cell's critical functions are accomplished by complex assemblies whose structures and conformational changes are best revealed by electron microscopy. Advances in sample preparation, instrumentation and image processing have supported visualization at increasing resolution, though usually insufficient for independent structure determination. Here, computer methods will be developed to take full advantage of known high resolution component structures - to elucidate molecular interactions between subunits, structural changes upon assembly and conformational transitions between functional states. Methods are already developed for the initial docking of known component structures into assembly density. Here, the focus is on refinement of more detailed flexible models. The foundations will be investigated first, such as the granularity and parameterization of models appropriate at different resolutions. Our development of model cross-validation methods allows these issues to be tackled objectively. Then stereochemical restraints will be evaluated to see which most effectively complements the available low resolution data. This work will include development of restraints for continuum electrostatics and hydrogen bonding, and a comparison of atomistic restraints with heuristics that might enhance convergence at low resolution. Finally, a variety of optimization algorithms will be evaluated, including least-squares, Monte Carlo and genetic algorithms. Guidelines will be developed to choose the refinement approaches best suited to the resolution range and experimental regime. The criteria will be maximization of model detail without over-fitting, as judged by the new cross-validation metrics. The best methods will be programmed in a software package to be distributed. They will be applied to electron microscopic studies of muscle protein and ribosome dynamics in collaborations with Ken Taylor and Joachim Frank respectively. Preliminary results demonstrate the potential for extraction of structural detail well beyond the nominal resolution limit, using restrained refinement. One goal is to probe the limits with more appropriate parameterization and restraint of the atomic refinement, enhancing generally the product of electron microscopy studies. A second goal is a more detailed mechanistic understanding of biomolecular complexes, going beyond domain movements to the internal conformational changes that drive them.